Dear Commons Community,
I have just finished reading a new book, Prediction Machines: The Simple Economics of Artificial Intelligence, written by Ajay Agrawal, Joshua Gans, and Avi Goldfarb, all affiliated with Toronto University’s Rothman School of Management. Published by Harvard Business Review Press, it attempts to lift the veil on the development of artificial intelligence (A.I.). The authors’ basic premise is that A.I. is fundamentally a prediction technology not necessarily an all consuming “take over” technology at least not for the foreseeable future. Regardless, its impact will be profound on much of our human endeavors and especially on the economic and business sectors. Here is an excerpt from a review that appeared in the New York Times:
“Compared with the amount of ink spilled over the prospects of artificial general intelligence and all its accompanying fears — the singularity! — there’s been much less attention to the smaller changes already happening in the realm of A.I. and their quite profound economic implications.
Enter “Prediction Machines,” which looks at just how far “narrow A.I.” has come over the past few years. Computers are already good at performing a single task for which they have been trained, making them more efficient and cost-effective than humans in many cases.
Of course, decision-making involves more than just being able to make accurate predictions, but A.I. is also being drafted for higher-level functions including using predictions to weigh outcomes and pass judgment.
For all their gains, though, computers are significantly better under certain conditions — say, when they have a lot of past data — and decidedly weaker in others, like predicting “unknown unknowns.” “Prediction Machines” does a good job of showing where computers work best and where humans still have an edge.
The authors argue, though, that we shouldn’t see this as an either/or fight to the death. In many cases, the best answer is to combine the powerful pattern recognition of a computer with the insight of a trained human.
Take one example they offer, from the field of medicine. A well-trained algorithm was able to find a certain type of breast cancer with 92.5 percent accuracy. Human pathologists were able to do so at 96.6 percent. Stop there and you would say that computers are getting quite good, but not quite as good as highly skilled humans, at least at this task.
But one need not stop there, and thankfully the researchers didn’t. Combining the work of computers and humans resulted in 99.5 percent accuracy. In part, that’s because humans and computers made different kinds of mistakes. It’s certainly a happier outcome to imagine that we and the machines could work together.”
The final chapter examines three trade-offs that are already being seen with this “infant” technology.
- Productivity versus distribution. A.I. will enhance productivity and create more wealth, however, it might exacerbate the distribution of wealth and lead to greater income inequality.
- Innovation versus competition. Businesses have incentives to build A.I. but this may lead to monopolization. Faster innovation may benefit from a short term perspective but not necessarily for the long-term perspective.
- Performance versus privacy. A.I. performs best with more data including personal data. The need for more and more personal data will come at the expense of privacy.
For all three of the trade-offs, the authors put forward that societies and jurisdictions “will have to weigh both sides of the trade and design policies that are most aligned with their overall strategy and the preferences of their citizenry”.
Good read for those interested in learning more about A.I.
Tony